Humanoid robot navigation and obstacle avoidance in unknown environments

A multi-gait approach for humanoid navigation in cluttered environments

“…A side-stepping gait can be easily achieved on the NAO robot simply by prescribing a zero forward speed and a non-zero sideways speed to the built-in walking routine. In [32], a path planning and obstacle avoidance system was developed and experimentally demonstrated on the NAO humanoid robot based on our GODZILA algorithm [28][29][30][31] utilizing a pair of light-weight, inexpensive, and compact ultrasonic sensors that are integrated into the NAO. A sample experimental run is shown in Figure 4.…”

“…The two additional components in GODZILA in addition to the optimization-based computation of the optimal heading address local motion to the target (local straight-line planner that provides a reference trajectory along which navigation is performed using the optimizationbased approach described above) utilized if the target is visible and navigation towards a random target to address possible limit cycle oscillations (while the introduction of the inertial term in the optimization cost has the effect of increasing the spatial extent of possible limit cycles, limit cycle oscillations can still occur if the obstacle set is spatially large since the GODZILA algorithm is based on local sensor data and no map of the environment is built; such potential limit cycles are addressed by the random navigation component of GODZILA). The implementation of the GODZILA algorithm on the NAO robot in [32] is based on the two onboard ultrasonic sensors (sonars) mounted in NAO's chest (transmitters oriented 40 degrees apart; receivers 50 degrees apart). While ultrasonic sensors are relatively inexpensive and compact, the spatial obstacle information that is obtained from these sensors is relatively low-resolution in multiple senses: a wide sonar beam within which there is no angular information as to where the sensed return is from, low range measurement accuracy (typically a few cm), and relatively small range (typically a few meters).…”

“…In particular, the ultrasonic sensors on the NAO robot have a resolution of 1 cm, a sensing range between 0.25 m to 2.55 m (i.e., an obstacle which is less 0.25 m away results in a sensor measurement of 0.25 m), and a viewing cone of about 60 degrees. While the compact ultrasonic sensors are associated with a high degree of uncertainty about the obstacles' location within the viewing cone as well as noisy/intermittent sensory characteristics at larger oblique angles and close proximity to obstacles [32], it was shown in [32] that an obstacle avoidance system can be implemented with these ultrasonic sensors in typical office-like environments through a "virtual" sensor concept to address these characteristics of the ultrasonic sensors. The obstacle range measurements/estimates for the GODZILA algorithm are generated as a combination of actual and virtual sensor measurements including intra-beam virtual measurements (to model the wide beam angle of the sensors with decreasing confidence estimates for larger offsets from the center of the sensor field of view) and prior-time virtual measurements (sensor measurements from a few prior sampling instants mapped to the current robot-relative frame by utilizing estimates of relative translations and rotations over successive sampling instants within a sliding window of time).…”

“…In particular, one approach that has been specifically designed to be computationally light-weight so as to address tight memory and processing contraints on small autonomous vehicles is the GODZILA algorithm [28], that enables navigation in completely a priori unknown environments without requiring building of an obstacle map. GODZILA is applicable to various types of unmanned vehicles [29][30][31] as well as humanoid robots [32]. In this paper, the autonomous navigation capabilities in cluttered environments that can be enabled by the human-like gaiting capabilities of biped humanoid robots are addressed.…”

“…In this paper, the autonomous navigation capabilities in cluttered environments that can be enabled by the human-like gaiting capabilities of biped humanoid robots are addressed. In particular, a new low-profile crawling gait is developed and the GODZILA based humanoid path planning and obstacle avoidance system developed in our previous paper [32] is extended to address the multi-gait humanoid motion capabilities including environment-dependent switching between a set of gaits including the new low-profile crawling gait in addition to two typical humanoid gaits that include forward walking and side-stepping motion. The low-profile crawling motions considered in this paper enable a humanoid to pass through very low openings in environments where walking or even crawling on knees (e.g., [12,13]) would not be sufficient to clear the minimum height requirement.…”